Condition-based maintenance, predictive maintenance, and reliability-centered maintenance get used interchangeably, but they solve different problems: CBM is a threshold-triggered action, PdM is a trend-driven forecast, and RCM is the decision framework that selects between them [S1][S3].
Across a rotating-equipment fleet the practical gap matters: a CBM program fires when a vibration sensor crosses a setpoint, a PdM program fits a degradation curve to estimate remaining useful life, and an RCM analysis decides whether either approach is worth the cost for a given failure mode in the first place [S3][S5].
What each strategy actually does
CBM relies on real-time sensor data (vibration, temperature, pressure, oil debris, acoustic emission) and triggers work only when a reading crosses a pre-set threshold against a known good baseline [S2][S5]. Fluke describes it as monitoring asset health in real time so crews can identify faults and repair them right away [S4]. The data inputs are typically single-point or limited-multipoint, and the action is reactive within a healthy window, not forecast-driven [S2].
Predictive maintenance aggregates multiple sensor streams (pressure delta, vibration, flow, current, lubricant particle count) and fits a trend to estimate when the asset will cross a failure threshold, so work can be scheduled into a planned outage rather than triggered at the alarm [S2][S7]. Upkeep's June 2026 framing is direct: predictive maintenance is more exact but more data-intensive, while condition-based is less exact but simpler to deploy [S8].
Reliability-centered maintenance is not a sensing technology. It is the structured decision process, formalised under SAE JA1011, that maps each failure mode to the right tactic: predictive, preventive, run-to-failure, or redesign [S3][S9]. RCM decides what to do; CBM and PdM execute the decision where it applies [S3].
Decision criteria: cost, complexity, failure-mode fit
The three approaches line up on four engineering decision criteria as follows. Capital and integration cost: CBM is the lowest, typically handheld vibration pens, infrared thermometers, and basic pressure transmitters tied to a SCADA alarm; PdM is medium-to-high, requiring permanently mounted vibration condition monitoring plus a historian and analytics; RCM is mostly analyst time plus the data infrastructure CBM/PdM already produce [S1][S6][S10]. Data intensity: CBM uses a single threshold, PdM uses trend windows across multiple parameters, RCM uses failure-mode and criticality data that often sits in a CMMS or risk register [S2][S7][S9].
Failure-mode coverage: CBM and PdM only work when a measurable warning precedes failure, so they cannot cover no-warning failures such as a blown fuse, a snapped belt, or a sudden bearing race fracture in a lightly loaded machine; RCM explicitly routes those failure modes to run-to-failure or hard-time replacement [S3]. Time horizon: CBM acts at the alarm, PdM plans weeks to months ahead, RCM sets the multi-year maintenance philosophy per asset class [S5][S9].
Fiix's September 2025 piece calls CBM and PdM a spectrum rather than a binary choice, with PdM positioned as the more capable but more expensive end [S6]. Augury's March 2026 implementation guide frames RCM as the layer that decides where on that spectrum each asset sits, citing risk-based maintenance (RBM) as the upstream prioritisation step that feeds RCM with the asset-criticality list [S9].
Where the savings actually come from
Vidyatec's October 2024 review cites the US CBM deployment figure of approximately $35 billion in annual savings at widespread adoption, drawing on the "On condition-based maintenance policy" literature [S1]. The same source cites a Plant Engineering survey showing 41% of industries had adopted predictive methodologies while only 22% used RCM as their main strategy, evidence that sensing investment has outpaced the decision framework meant to allocate it [S1].
MaxGrip's April 2022 piece, still the canonical process-industry comparison, lists CBM's three concrete cost levers: avoiding work on healthy non-critical assets, minimising downtime by acting before failure, and shifting work into planned outage windows [S2]. PdM extends the same logic by replacing the alarm-trigger with a forecast, so spares, labour, and permits can be staged in advance instead of rushed [S2][S4].
Faraday Predictive's November 2024 update frames the RCM business case around the failures you choose not to instrument: in cases where detection is technically possible but uneconomic, or where the failure is too fast to respond to, RCM directs spend away from CBM/PdM and toward redesign, redundant equipment, or scheduled replacement [S3]. That is where the second wave of savings tends to come from, after the easy alarm-tuning wins are exhausted.
When NOT to use each strategy
CBM is the wrong tool when failure gives no measurable precursor, when the cost of the monitoring hardware exceeds the avoided-loss value, or when the only available "signal" is a hand-held reading that depends on operator interpretation [S2][S3]. MaxGrip flags exactly this: CBM is simple, but it relies on human factors for data collection and reading accuracy, so inconsistent information is the main limitation [S2].
Predictive maintenance is the wrong tool when the asset class lacks historical failure data to fit a trend, when the failure mode is dominated by random events rather than degradation, or when the plant cannot sustain the data-pipeline engineering that trend analytics require [S6][S8]. Upkeep's 2026 update warns that PdM is "more exact but more intensive," a poor fit for small fleets where a single reliability engineer has to cover everything [S8].
RCM is the wrong tool as a first move on a plant with no baseline condition monitoring system and no failure history; the inputs simply do not exist yet, so the analysis becomes opinion dressed up as a decision matrix [S3][S9]. Oxmaint's July 2025 integration write-up explicitly recommends layering RCM on top of an existing preventive and CBM program, not standing it up on its own [S10].
Putting it together: a per-asset decision rule
For a critical rotating asset with clear degradation precursors (pump bearings, gearbox wear, motor winding temperature), the rule of thumb from the literature is: install permanent vibration condition monitoring and pressure sensor channels, run PdM on the resulting trends, and use RCM only to validate the criticality ranking and decide on inspection intervals [S3][S6][S9]. For a non-critical utility asset with no measurable precursor, RCM should explicitly route to run-to-failure or scheduled replacement, and any CBM/PdM spend on it is wasted [S3].
For the middle band, secondary pumps, fans, HVAC motors, the flow meter and industrial valve populations, Nuvolo's three-way framing (preventive is time-based, CBM is threshold-based, PdM is trend-based) maps cleanly onto instrumentation cost: portable CBM instruments are the right starting point, with permanent PdM sensors added only where the failure cost justifies the install [S5].
Two trackable signals to watch through 2027: published adoption rates of risk-based maintenance (RBM) as the upstream layer feeding RCM analyses, and integration of RCM outputs into CMMS work-order generation rather than as a standalone document [S9][S10]. The Augury March 2026 implementation guide treats that integration as the main open problem, and the Oxmaint July 2025 piece treats it as the main payoff, which is a useful tension to follow [S9][S10].
For assets that operate under heavy shock and vibration, a hermetically sealed vs epoxy sealed accelerometer decision directly shapes the signal quality a PdM trend algorithm sees, and that signal quality is what separates a usable CBM threshold from a noisy one.